her of Caisse d'Epargne, France's third-largest retail ba
nk.
Like so many other industry buzzwords, the term
knowledge management
means different things to different people. Basically, it's a business philosophy that encourages companies to use intellectual assets to their fullest. In computer science, knowledge management (KM) refers to technologies that intelligently analyze similar business cases, extract and classify patterns, and autonomously gather information according to user preferences.
Case-Based Reasoning
Several European developers such as Acknosoft, ISoft, and TecInno are now offering commercial tools for building case-based-reasoning (CBR) systems, which is probably the most common KM application in use today. CBR serves as a problem-solving engine that judges a new problem
by comparing
it to already-classified cases. Companies are using CBR applications in industrial design, planning, and explanation systems, where they help find solutions to similar recurrent problems. Other promine
nt CBR application areas are call centers and help desks for technical support as well as corporate memories.
A CBR system allows for fuzzy queries. With CBR, you do not specify exact criteria to be matched. Instead, you enter a new case as the basis of the search. The CBR retrieval engine exploits the complete description of the case and computes a percentage of similarity for each variable describing the case. Based on this level of similarity, it identifies comparable cases. Thus, CBR finds cases that traditional engines with exact search criteria miss.
To create a CBR system, you do not need to understand how to solve the problem. You need only records or cases of previously solved problems. Consequently, CBR significantly reduces the knowledge-elicitation process from subject-matter experts.
However, the biggest challenge in CBR is maintaining and organizing a comprehensive base of classified cases. For example, CBR systems developers must decide whether it is more appropriate to store
information about sales revenue and sales representatives separately or to store just the representative's percentage of total sales. Therefore, to organize the case base, developers still need the detailed input of domain experts.
Rule Engines
Rule-based systems are currently moving out of the classical expert-system domain into broader application areas. They help model business practices in work-flow applications,
aggregate information
from electronic sources into a corporate knowledge base, and automate document compilation according to customer specifications. The last two applications are gaining broader acceptance because of the increased use of the Web. New rule-authoring tools, such as Neuron Data's Jewel, that are based on Java are starting to address this market.
The better organized the case base is and the more comprehensive it is, the better the CBR system will be at solving new problems. Rule-based systems, on the other hand, require the developer to s
olve the problem in advance and structure this knowledge into fixed rules.
"CBR systems learn; rule-based systems usually do not," says Ian Watson of the University of Salford, U.K. When a corporate environment changes, system administrators must elicit new rules from experts and add them to the rule base. Watson calls this the "knowledge-elicitation bottleneck," a disadvantage of rule-based systems. But the latest rule-based development tools use pattern- matching techniques to alleviate the rule-elicitation process.
Manufacturing company Baumer Electric (Frauenfeld, Switzerland) implemented a rule-based application that stores the if-then decisions engineers have to make when creating a parts list for new orders. It automatically creates a parts list in 15 to 45 seconds. Previously, engineers needed anywhere from 15 minutes to several hours, according to the company. Moreover, the application has access to the supply database and notifies users when parts are not available, how much they cost, a
nd when they can be delivered.
Intelligent Help Desks
Most help-desk applications typically use elements of KM technologies: fuzzy searches, decision trees (see "Data Mining at Your Desk," July BYTE International Edition), and CBR. Aimed at support specialists, help-desk systems aid them in finding and ranking past cases pertinent to a current technical problem.
Compared to previous help-desk systems, CBR-based applications can reply automatically to requests. When combined with groupware or e-mail, it is possible to build an automatic in-house support center.
For example, the airline telecommunications and information services provider SITA (Nice, France) integrated ISoft's CBR toolkit into Lotus Notes to automate responses to user requests due to network problems or outages. If a user needs help, he or she fills out a request form and sends it to the network manager. The automated help desk then translates the literal description into a problem case, kicks off the CBR process, an
d returns a number of solutions in the user's language. If the system-generated instructions do not solve the problem and the inquiry comes back, it goes directly to the network administrator.
One of CBR's strengths is its ability to learn and to automate the mundane and repetitive part of an expert's work. But a CBR system can also support complex decisions without the interrogation of a human expert.
A good example is Caisse d'Epargne, which developed an expert system that helps find the best locations to place new ATMs. Caisse d'Epargne receives more than 30 requests a month from its retail banks to set up new outlets all over the country. In 1995, it started to build a database of requests that now contains 1500 entries. Every case contains a description of an existing ATM, with about 200 attributes.
When the bank's domain experts now receive a request for a new ATM, they enter the specifics into the system. The system returns a list of similar installations ranked by how much in common
the new site has to the others. This way, they can more accurately predict profitability of new applications and make decisions much faster. According to the Caisse d'Epargne ATM-installation team, its expert system accepted or rejected 80 percent of the sites without human intervention. The bank reckons it made its return on investment within a few months, because it was able to avoid the installation of 10 ATMs that would have proven to be unprofitable.
However, not all KM implementations end successfully. Early this year, a Swiss retail bank put on hold indefinitely an advanced knowledge-based system that was to offer automated 24-hour, seven-day-a-week customer services to clients over the Internet. The project leader says that the KM aspects worked out fine, but they couldn't get a handle on the data-organization problem. He says that the databases that were to feed the application were incomplete, too diverse, and incompatible with each other.
At Unilever PLC (London, U.K.), a big player in
the food, chemical, and personal-health product sectors, Autonomy's agents let companies "cycle" or distribute knowledge throughout the entire organization. By profiling both overtly and covertly the areas of interest of knowledge workers, the system, known as I3, can automatically index information displayed in a user's Web browser. Whenever I3 detects new postings or documents on the intranet that are of particular interest, it alerts the user. For example, if someone posts a corporate memo about the results of a customer survey, I3 immediately informs all interested Unilever employees via e-mail, a pop-up window on the screen, or even a voice message on the person's answering machine. Says Autonomy's Domenic Johnson, "This kind of knowledge sharing avoids reinventing the wheel."
Living Up to the Hype
KM systems not only help a company to use existing resources more efficiently but also to share worker experiences across an enterprise. Corporate memories are a constantly updated and highly st
ructured database of business practices. So far, they have been used mainly in the health-care industry. They assist in medical diagnosis and also improve patient care by offering lessons learned from past cases.
For example, the consultancy CIBIT (Utrecht, The Netherlands) has a health-care application that consolidates the knowledge shared by medical workers treating geriatric patients across the country. Rob van der Spek, strategy and development manager with CIBIT, says, "It acts as a national collective memory." Users enter a case and search the repository for similar cases to determine the best procedure, indexed by such factors as care period, care plan, and patient satisfaction. It also includes an induction mechanism used to identify the most pertinent search criteria. Workers simply enter the new case details, and the system suggests diagnosis and patient-care advice adapted from previous cases.
Because so many software vendors and business consultants are praising KM, it may have proble
ms living up to all its hype. When it works it's wonderful, saving time and money, opening up new vistas of information sharing, and automating mundane parts of the knowledge worker's work load. However, setting up a successful KM system can be a complex process, especially if vital information is spread far and wide throughout disparate databases in a large organization. Nevertheless, KM tools are entering the mainstream. Simply put, a well-implemented KM system has the power to unleash a corporation's global knowledge assets.
Where to Find
Acknosoft
Paris, France
Phone: +33 1 44 24 88 00
Fax: +33 1 44 24 88 66
E-mail:
manago@ibp.fr
Autonomy Systems
Cambridge, U.K.
Phone: +44 1223 421 220
Fax: +44 1223 421 583
Internet:
http://www.agentware.com
Corporate Memories
